Insight
AI Impact on Design Roles
"Learn to design with AI" gets handed around like it's one skill to pick up. It's four separate crafts — using the tools, designing the collaboration, designing for the agents, and shaping the model itself — and only the first is turning into something every designer is simply expected to do.
The one skill everyone means
What most people mean by "designing with AI" is a single skill: a designer driving the tools to produce the work faster. Five layouts, a clickable prototype, the docs — all in an afternoon. One new rung on the old ladder.
Actually four crafts
The phrase hides four different jobs, each with a different user: designing with the model to produce the work, designing the product where a person and the model collaborate, designing for the agents that call your software, and designing the model's own behavior.
The waterline rises
Driving the tools is sinking below the line into table stakes — expected of everyone, a differentiator for no one. What lifts a designer above it is the other three: shaping AI-native products, the systems agents read, and the model's own behavior.
A few years ago the job had a clear center: you made the screen. A producer needed a renewals dashboard, so you spent two or three days in Figma — drawing the tables, laying out the flow, wiring a prototype — and the craft was in how well you produced that artifact. "Designing with AI" showed up as an upgrade to exactly that. The same job, done faster. Generate five dashboards instead of drawing one.
But "faster at making screens" is the smallest thing AI did to the work. Underneath that one phrase sit four different crafts. There's designing with the model — the tool skill, the one everyone means. There's designing the product where the user works alongside the model, choreographing a back-and-forth instead of laying out a screen. There's designing for the model, when the user isn't a person at all but an agent calling your software, and the interface becomes a contract. And there's designing the model — deciding which documents it prioritizes, when it flags uncertainty, when it asks the broker and when it just proceeds.
Only the first of those is becoming table stakes. Driving the tools is turning into something every designer is expected to do, the way knowing Figma was — a floor, not an edge. The other three are where the role is actually splitting. They ask a designer to shape products, systems, and behavior that didn't exist to be designed a few years ago, and most of the field hasn't caught up yet. That gap is the whole opportunity.
You can watch design teams sort themselves along the same line. The ones treating AI as a faster Figma stay where they were, producing interfaces a little quicker. The ones asking how a person should collaborate with a model, how an agent should read their product, how the model itself should behave — those are the ones designing what comes next.
Producing interfaces is the floor now. Shaping the AI is the job.